Proof over Promise Insights on Real World AI Adoption from 2025 MINDS Organizations 2026

SPOTLIGHT 9 Hyundai and DEEPX optimize for real-time edge computing Hyundai and DEEPX are redefining AI infrastructure for robotics by integrating custom AI semiconductors for hardware acceleration with proprietary optimization software. This approach enables real-time inference on compact, battery-powered delivery robots without relying on cloud connectivity or energy-intensive graphics processing units (GPUs). Their chips deliver GPU-level performance while reducing power consumption and heat generation, critical for indoor environments with strict energy constraints. Supported by development tools for building applications, the solution accelerates deployment and scalability across edge applications. Impact: Hyundai and DEEPX’s patented AI hardware and software bring real-time intelligence to edge devices with 70% less power. This enables affordable, scalable automation for robotics and internet of things (IoT), improving operational efficiency and presenting new opportunities for innovation. Analysis of all applicants to the MINDS programme shows that, while cloud adoption has unlocked scalability and innovation, organizations continue to navigate a dynamic balance between on-premises and cloud investments. Competing priorities such as cost, standardization, flexibility, sovereignty and risk diversification are driving ongoing trade-offs. As such, organizations are investing in a range of AI infrastructure modernization strategies: –Overall, 55% of all applicants to MINDS represented hybrid architectures, blending on-premises and cloud capacity to balance control, flexibility and scalability. This enabled organizations to handle multi-domain workloads, expand R&D capacity and integrate diverse AI tools across functions, supporting complex use cases in larger organizations. –In total, 15% of applicants are anchoring on-premises computing infrastructure in environments where sovereignty and data ownership dominate, including in-house LLMs, or where raw performance is critical, e.g. simulation-heavy R&D. –Additionally, 30% of applicants are pursuing a cloud-first infrastructure strategy for flexibility, speed, global reach and instant access to cutting-edge AI services. Example workloads include marketing analytics and mainstream software development, where agility represents a key advantage. –Edge computing served as an add-on capability for 18% of all applicants, bringing inference to where data is generated to achieve real-time responsiveness and energy-aware scale in distributed settings such as smart manufacturing cells, robotics lines and dense IoT networks. –For the heaviest computing power lifts, 5% of applicants turned to high-performance computing to underwrite model training and physics-grade simulations. Proof over Promise: Insights on Real-World AI Adoption from 2025 MINDS Organizations 22